Comparative Performance Analysis of Machine Learning Classifications for the Prediction of Accident Severity on Motorways (Freeway) of Pakistan

Abstract

The research thesis focuses on the prediction of accidents severity on the Motorways of Pakistan using machine learning classifiers. With traffic accidents being a significant global issue causing numerous casualties and economic losses, the objective is to compare and identify the best machine learning classifier for accurate severity prediction. Accidents data from 2010 to 2020 on Motorway M-2 were obtained from the National Highway and Motorway Police (NH&MP). The dataset was preprocessed by discarding irrelevant attributes and converting it to nominal values. KNIME, an open-source software, was utilized to train and test various machine learning classifiers including Random Forest (RF), Logistic Regression (LR), Support Vector Machine (SVM), Gradient Boosting (GB), Decision Tree (DT), ANN (MLP), and ANN (PNN). The performance of these models was evaluated using a 70% training and 30% testing data split. The results revealed that the Random Forest RF model outperformed other classifiers, achieving an accuracy of 94.60%, precision of 93.60%, and an F-1 score of 0.938. The findings highlight the potential of the RF model for accurate accidents severity prediction compared to Logistic Regression (accuracy: 84.90%) and Naive Bayes (accuracy: 86.50%). This research provides a foundation for extending the analysis to other road networks and datasets, contributing to the improvement of road safety measures.